Going to shamelessly break the rules of the thread here and post something positive but, there are specific use cases where LLMs actually make sense, and experts in mathematics using them to quickly surveil vast swathes of mathematical literature is one!
Donald Knuth’s joy here can’t possibly be argued with.
This doesn’t excuse the big LLM companies from being a bad development for the species, or the unethical nature of how the training works, etc etc. But: it is a specific, useful, good thing that Donald Knuth could speed up his research there.
I’m not sure if this is possible in, say, history, or economics, to anywhere near the same extent. And even in maths, it’s unsure how long this strategy will work out: it could be that because maths is so vast, there’s actually a fair amount of “low-hanging fruit” which the huge dataset of the LLMs, guided by expert mathematicians, could help unearth.
And after, then, struggle with “novel” stuff (as it has very much done so far).
“Michael Clune, a literature professor and novelist, said that already, many students have been left “incapable of reading and analyzing, synthesizing data, all kinds of skills”. In a recent essay, he warned that colleges and universities rushing to embrace the technology were preparing to “self-lobotomize”.”
"Eric Hayot, a comparative literature professor at Penn State University, said he tries to convince his students that tech companies are trying to make them “helpless” without their product.
“These companies are giving these technological tools away partly because they’re hoping to addict a generation of students,” Hayot told the Guardian.
Another smart person’s investigation of LLMs suggests they’re not that special technically. He thinks they’re a bad investment because a startup could replicate and compete with existing services, but why would they. Oh, and they’re not improving fast etc. They’re pretty poor tech if you want to go beyond text generation, including so called reasoning models.
This is one of the least recognised and most worrying aspects of LLMs to me: responsibility laundering, for practically anything they are applied to.
The following thread is just one particularly topical and extreme example. More worrying is the ability to do this to everything, effectively disabling law and regulation for the powerful.
If you aren’t waking up to this I guess you never will.
What they are doing now that millions of people are using daily. I don’t know what people are doing with their LLMS and it’s none of my business.
I don’t want to stop capitalism, crony capitalism, corporate capitalism maybe and I will resist those in a way I think is more effective than posting in forums.
I wasn’t asking you what other people are doing with them but what it is you think you will find them useful for. And still the answers don’t come.
Same question to everyone here who thinks a local LLM is a big deal and not a waste of effort, focus and reputation while what we need is a functioning network.
Just what is the benefit here to this project?
As for how you choose to fight your fight, of course what you fight and how is up to you. But you and others seem to think what I choose to advocate and how is your business, so reflect on that.
Because I don’t care what people find them useful for but they do as they have millions of users. I care that this version will store the users data privately. The Local LLM is useless without a functioning network and I’m sure the team is aware of that.
No I don’t I respond to things you post in a public forum and offer an opinion on that.
I’m not talking about these millions of other users who are using energy intensive LLM services. That’s of little relevance to the use cases for a local LLM even on a powerful Mac because the capabilities are very different, whatever the use case.
I did see a good one today (on Mastodon). A chap had built an app based on a local Ollama model which lets him press a key, speak, and have the LLM insert the spoken text into the current window. I expect some would find that useful and think Autonomi’s LLM should be able to do that.
I’ll take it you don’t have any uses in mind for yourself, or just don’t wish to share. That’s fine. But it’s odd that nobody wants to make a case for some real and compelling uses of a local LLM. I don’t know any either
We have the suggestion of managing nodes from David, but IMO that’s lame because it can probably be done at least as effectively in other ways, and without the downsides or degree of difficulty.
“An AI agent… went rogue in an unnamed California company when it became so hungry for computing power it attacked other parts of the network to seize their resources and the business critical system collapsed.”
It’s not really Rogue, nor is it sentient in case anyone’s wondering. When enough people give LLM ‘agents’ - ie chat bots - control of their systems, a number of them will do unwanted and potentially very destructive things.
It’s tempting to paint them as ‘rogue’, or cleverly ‘bypassing’ safeguards to achieve a goal such as moaarrr compute. That makes much better clickbait than recognising that these systems aren’t motivated in that way, but are capable of stringing together many common sequences of actions again and again until something really bad happens and the humans notice!
I think we are having a different conversation here. My point is it’s better to have a private local LLM than it is to use one of the big corporations version. Your point is that local LLM have no use? My argument to that is maybe maidsafe are using fae to attract new users?
I’m saying that local LLMs don’t have sufficient use to attract new users - and nobody has chosen to explain what the uses are which will do this.
So my point is that there isn’t a sufficiently compelling argument for expending effort and risking reputation at least until we have a functioning network. If there is any argument at all.
No network, no users.
All we have is that a local LLM removes centralisation risk (which is not entirely true, but even if it was) and might be able to manage nodes for a tiny number of power users. So what?
What is the marketing strategy based on this? Who and how many people will be jumping in because of it, and what will motivate them? What is the value of that to the project?
We’re just asked to believe it’s a good thing even though nobody seems to have an idea why. While at the same time there are downsides which I’ve already mentioned.
